The predictive skill of a simple model of surface flow on the Scotian Shelf is assessed using oceanographic data collected in February 1996. The model is forced by wind stress, water density, and sea level along the open boundaries of the model domain. The skill of the model with respect to subtidal variations of alongshore current and bottom pressure is quantified by the ratio γ2 = Var(O - P)/Var(O), where Var(O - P) and Var(O) are the variance of the prediction errors and observations, respectively. Skill is highest for bottom pressure (γ2 = 0.2) followed, in order, by cross-shelf gradients in bottom pressure (γ2 = 0.5), horizontally averaged currents (0.4 < γ2 < 0.7), and currents at individual current meters (0.6 < γ2 < 1.4). The skill of the model with respect to drifter position is quantified by the search radius, centered on the predicted drifter position, that ensures a 50% chance of locating the drifter. Skill varies significantly with time but is generally highest when the drifter motion is strongest. We conclude with a comparison of the performance of the model against predictions based on the release point of the drifters and the operational scheme presently used to guide marine search and rescue in this region.